A recent analysis highlights an aggravating regression in newer Claude models regarding their ability to reliably execute tool calls. Despite overall improvements in general intelligence and reasoning, this specific capability has significantly declined. This downgrade is causing major frustrations for developers who depend on the AI for consistent external integrations.